A Moment That Stopped Me
I follow a knowledge community with tens of thousands of members, run by someone who has been producing content for many years. Its tagline is a single sentence:
In the information age, information is cheap. What’s valuable is the ability to process it.
I’d heard that sentence before. It sounded right, but it didn’t really register — until AI actually entered my daily workflow, and for the first time it became concrete enough to unsettle me.
Before AI, “having information” was still, to some extent, a moat: you subscribed to sources others couldn’t reach, you’d read books others hadn’t, you had a hard drive full of material nobody else could find — and all of that translated into some kind of edge.
Now, a model can generate ten thousand words in a second. It can bookmark, summarize, translate, and retrieve almost anything for you. The barrier to “having information” has been cut to zero. The only thing that remains scarce is the whole pipeline that turns information into capability, and capability into finished work.
So I looked back at my own note-taking system and found an awkward truth: I’ve been taking notes for years, my bookmarks number in the thousands, I’ve switched note apps five or six times — and none of it has made me meaningfully stronger. Most of what I’ve stored has never been called on again.
Where did it go wrong? I thought about it for a long time, and the conclusion was: I had been treating four completely different things as if they were one.
Four Things, Four Stages
I now split them into four layers: information, records, knowledge, creation.
Information (raw / external noise)
│ capture, noise reduction
▼
Records (index / internal semi-finished)
│ structuring
▼
Knowledge (schema / reusable)
│ recombination for an audience
▼
Creation (finished product)
In everyday language we blur these four words together almost constantly: we treat bookmarking as learning, note-taking as mastery, and notes written for ourselves as content ready to publish. But the moment you separate them, a lot of long-standing friction suddenly makes sense.
Information is raw input — meant for AI to process, or for you to skim quickly. Most of it is noise. It might be something you gathered yourself, something AI generated, or a signal that’s currently scarce in the market. At this layer, the only thing you should do is capture and reduce noise — expose yourself to as many high-quality sources as possible, while ruthlessly keeping noise out. The test isn’t “is it useful,” but “is it worth moving into the next stage of processing.”
Knowledge is structured, reusable, and relevant to you personally — something you can call on again and again. A mental model, a handful of skills, some tools and methodologies, and your own identity, positioning, values, and judgment. The keyword for knowledge is reusability: only once something is structured into a form you can repeatedly reuse does it deserve to be called knowledge. It solves your own problems.
Creation is the finished product, recombined for a specific audience. It corresponds to a platform’s recommendation logic, a particular group of users’ reading habits, and the substantial research you did to support this specific piece of expression. Creation solves other people’s problems — the goal is to have your audience receive it, understand it, and want to connect with you.
Knowledge and creation, you’ll notice, are fundamentally different things: one faces inward, the other faces outward; one asks “can I reuse this,” the other asks “can others receive this.” Trying to create the way you build knowledge produces self-indulgent writing only you can understand. Taking notes the way you create wastes huge amounts of energy making things “look good” without ever building any real capability.
The Layer Everyone Overlooks: Records
Of the four layers, the one most commonly misunderstood is records.
Most people only recognize three tiers: information → knowledge → creation. Records get quietly filed under “knowledge.” But I’ve increasingly come to believe records deserve to stand alone, because they are an independent intermediate form.
A record is essentially an index. It’s relevant to you, but not necessarily useful forever — it might just be something you’ll need someday, or something you’re using right now to clarify your own thinking. That kind of thing doesn’t yet qualify as knowledge, because only what’s structured for repeated future reuse counts as knowledge. But records are still extremely important, because they are one of the highest-conversion methods for turning information into knowledge.
In other words: a record is semi-finished knowledge.
This explains a common illusion. Many people believe they’ve “learned a lot,” when really they’ve just “recorded a lot” — they stopped at the index layer and mistook the semi-finished product for the finished one. Bookmarking ≠ recording. Recording ≠ mastery. Information has merely been moved into your warehouse; it hasn’t gone through any processing by your own machine.
Someone in that community wrote a retrospective that nailed this precisely: the biggest misconception about a knowledge base is treating it like a bookmark folder — anything that seems valuable gets bookmarked, downloaded, saved, dumped in — and the result is that the knowledge base keeps growing while what you can actually call on keeps shrinking. The real problem was never too little knowledge. It was too much unprocessed information.
This is exactly why I insist on separating records into their own layer: it’s the first site where you personally process information with your own hands. The moment you write something down, you’re forced to choose, to compress, to restate — and those three actions are precisely where the path from information to knowledge begins.
The Easiest Mistake to Make in the AI Era
The framework is laid out. Now the real question begins: once AI enters the picture, how should this pipeline actually run?
Most people’s first instinct is to outsource the entire pipeline to AI — let it capture for me, summarize for me, write for me. It sounds appealing, but I’ve observed a recurring trap, and someone in the community described it perfectly in one of their retrospectives:
When I have a really enjoyable conversation with AI, it means I didn’t grow that day.
Why? Because what he gave the AI was all “direction,” never “task.” “Help me think through how to approach this topic” — that’s a direction. A direction has no completion state, so the conversation can go on forever, feeling more and more pleasant, and yet when you stop, you realize nothing remains. Real growth is always accompanied by friction and discomfort, and he had been using AI to eliminate friction — but friction is exactly what he needed most.
Mapped onto our four-layer framework, this becomes very clear: AI is best at the information layer, and it’s precisely the knowledge layer where it should least be doing the work for you.
Capturing, reducing noise, summarizing, and doing initial organization on information — hand that to AI, and you get a tenfold efficiency gain. But the record and knowledge layers — compressing information into your own judgment, structuring it into a model you can reuse — once you outsource those, you’ve outsourced the very act of “getting stronger.” You end up with an increasingly capable AI, and an increasingly hollow version of yourself.
The correct usage that circulates in that community resonates with me: present yourself manually first, then use AI to discover the boundaries of your own thinking. You have to run through it yourself once, establish a baseline, before AI knows which direction to adjust toward; otherwise it can only wander alongside you. AI is here to assist your thinking, not to think for you.
So I made myself a rule: use AI generously at the information layer, and insist on going through the knowledge layer by hand first. This one rule is the foundation for everything else in this series.
The Pipeline Isn’t a Straight Line
One more misconception worth clearing up: these four layers aren’t a straight line you’re required to walk in full.
It’s not that every piece of creation must start with “gather a hundred pieces of information,” nor that a bigger knowledge base means stronger creative output. The real path is often: a large amount of information plus a small amount of knowledge, recombined directly into a piece of creation. That information might be something you gathered on the spot, something AI generated in real time, or a signal that’s currently scarce in the market — while that “small amount of knowledge” is your real moat. It’s what determines how that pile of information gets cut, arranged, and pointed toward a conclusion.
So don’t be intimidated by “systems.” What matters was never how much you’ve stored, but whether real processing actually happened at each layer: was the information denoised, was the record structured, was the knowledge called on, did the creation actually reach someone. Ten thousand unprocessed pieces of information stored away are worth less than three knowledge cards you keep calling on.
What This Series Sets Out to Solve
With the framework established, all that’s left is to bring it down to earth, layer by layer. The next four essays in this column map exactly onto the four stages:
- The information layer — how to capture, how to reduce noise, how to distinguish “for AI,” “for a quick scan,” and “pure noise,” and how to hunt for scarce signal in an era when AI mass-produces content;
- The record layer — why “writing something down every day” is the highest-conversion action there is, how to let the semi-finished product settle with the lowest possible friction, and how a “next-day polish” uses a cooling-off period to push a record toward knowledge;
- The knowledge layer — how to transform a knowledge base from a bookmark folder into a “capability sediment zone,” how PARA and knowledge cards land in practice, why “your folder structure is a map for AI,” and the retirement mechanism behind “if a piece of knowledge goes unused, delete it”;
- The creation layer — how to recombine knowledge for an audience, how to build a content flywheel that runs “inspiration → processing → article → video → feedback → new insight,” and exactly where AI should stand at this layer.
On tools, I’ll get specific about note-taking stacks like Obsidian and Flomo, and how an AI agent like Claude threads through the pipeline. On cases, I’ll keep drawing from the real practice of that community of tens of thousands — it is itself a machine that turns information into capability and then into finished work, the best living sample of this entire framework.
Information is cheap. What’s valuable is the ability to process it. And “the ability to process information,” broken down, is exactly these four stages. Let’s take them one at a time.
This essay is the overview of the “From Information to Creation” column. Next: the first layer — Information: capture and noise reduction.
Responses